Google's AI is being manipulated. The search giant is quietly fighting back

In an experiment confirmed by the BBC, just a single article on a personal website was enough to make ChatGPT and Google AI Overviews answer that the author was a "world champion hot dog eating competitor."

The core problem is that AI search synthesizes multiple sources and presents them as a one-sentence definitive answer, making users less suspicious of the results than they would be with traditional search. This potential for manipulation was also confirmed on sensitive topics like health supplements and retirement/financial information.

Last week, Google updated its spam policies to explicitly define attempts to manipulate AI responses as a violation. Violating sites can be removed or have their rankings lowered in search results.

However, according to the article, manipulation may not end with blog spam. As crackdowns on blogs intensify, manipulators could move to more covert channels like YouTube influencers, and since Google's AI also cites YouTube, the cat-and-mouse chase is likely to continue.

Recently, Google and other AI products have also been experimenting with responses, such as more frequently attaching uncertainty labels to certain queries or advising users to check third-party reviews. However, experts believe such measures alone are not enough, and that confidently-given AI answers should still not be trusted without verification.